Voice dictation: option to disable profanity filter

Summary

Claude Code's voice dictation feature (hold-spacebar to speak) automatically censors profanity by replacing letters with asterisks (e.g. ku*wa instead of kurwa). There is currently no documented way to disable this filter.

Affected feature

Voice dictation as documented at https://code.claude.com/docs/en/voice-dictation — hold spacebar to dictate, transcription is inserted into the prompt.

Why this matters

1. Lossy transcription is wrong by default for a developer tool. The transcription should reflect what was said. Filtering changes the user's words without consent.
2. Especially impacts non-English users. Many languages use profanity in everyday speech without it being offensive (e.g. Polish kurwa is closer in register to English "damn"). Censoring it removes information from the prompt and creates visible artifacts in the model's input.
3. Inconsistency with the broader product. Claude itself does not refuse to produce profanity in code comments, commit messages, or chat. Censoring voice input is inconsistent with the rest of the product surface.
4. The CLI is a private developer tool. There is no broadcast audience, no child safety concern, no moderation context that justifies filtering by default.

Proposed solution

Add a toggle in settings.json:

{
  "voice": {
    "profanityFilter": false
  }
}

Default can stay true if Anthropic prefers, but the option must exist.

Alternatives that would also work:

• A /config toggle in the interactive CLI
• An environment variable like CLAUDE_CODE_VOICE_RAW=1
• A CLI flag

Workarounds attempted

• Searched settings.json schema — no relevant option exists.
• Searched /help and /config — nothing.
• Checked docs at https://code.claude.com/docs/en/voice-dictation — no mention.
• Searched existing issues — none on this topic.

Environment

• Claude Code CLI (latest), Windows 11, Polish locale.

Thanks.